Ten conversations you do not have yet
Lesson 8 of 24, level 01Index
You need to talk to real users. You have no research panel, no budget, and a founder who is nervous about you emailing customers.
Ask about the past, never about the future
Ask about last Tuesday
- Situation: When did it happen?
- Action: What did you do?
- Cost: What did that cost?
People are unreliable narrators of their future selves and reliable reporters of last Tuesday. So anchor every question in a specific recent event: tell me about the last time this happened, what did you do, what did it cost you.
“Would you use this?” gets a polite yes from everyone, including your mother. It is worthless. Ask about cost instead: what did that cost you in time, money, or a workaround you built? If the answer is “nothing really”, you have found a problem nobody will pay to solve.
The most valuable sentence in most interviews comes three seconds after you would normally start talking again. Count to five.
A good interview produces a reconstructable episode. Your job is to understand the participant's world before persuading them that your idea belongs in it.
Use your chosen course project throughout. The additional examples below are fictional practice cases; transfer the method to your own evidence.
Prepare a learning question
Start with the decision you need to make, then list what you do not know. Recruit someone who recently attempted the relevant task. A friendly person who has never faced the situation can help rehearse questions, but cannot validate the problem. Ask permission before recording and remove identifying details from shared notes.
Follow the sequence
Ask what happened last time, what they did next, what they used, and what the result cost them. Pause after an answer. A specific artifact such as a redacted spreadsheet often reveals more than a general opinion. Avoid introducing your solution until you understand the existing process and its constraints.
Debrief without smoothing
Separate direct quotes, observed behavior, and your interpretation immediately after the session. Capture contradictions rather than resolving them through storytelling. “Would use” is a statement about a possible future; a paid workaround is a past action. Neither alone tells you how common the problem is.
Follow one real episode
- Trigger: Why that day?
- Action: Show the steps
- Workaround: What filled the gap?
- Outcome: What happened next?
How The Mom Test handled it
The Mom Test
Rob Fitzpatrick's book opens with a founder asking his mother whether she would use his app. She says yes. She is lying, kindly.
The fix is to never ask about the future or about opinions, only about specific past behaviour. Facts about the past are hard to fake, predictions about the future are worthless from everyone, including you.
You cannot ask people to design your product. You can ask them what they did, and design from that.
Doing it with AI, and where it breaks
Draft your guide, then have a model audit every question for leading bias and for asking about the future rather than the past.
It passes questions that are subtly leading because they are grammatically neutral. “How important is speed to you?” presupposes speed matters. Ask specifically: which of these questions presuppose that the problem exists?
Your AI workbench
Start with your own notes or clearly labelled practice data. Remove private details before sharing. Replace the placeholders, run the prompt in your chosen AI tool, and keep the output beside its source.
Review this interview guide against my learning question: [question]. Flag leading, hypothetical, double-barrelled, and solution-revealing questions. Rewrite each as a question about a recent event. Return the original, the risk, and a replacement. Do not generate participant answers. Guide: [paste].
Before you use the output
- The guide asks about actual events
- The participant matches the task context
- Quotes and interpretations remain separate
Stuck? Try this next
Use AI as a rehearsal partner to spot awkward phrasing, but label all simulated answers synthetic. Never add a simulated interview to your evidence count.
Keep a brief AI log: input used, useful output, what you checked, and what you rejected. The decision remains yours.
Build it
A screener, an interview guide, and three recorded conversations.
Zoom or Google MeetOtter or Whisper
Checkpoint
If you did the build, these take two minutes. If you cannot answer one of them, that is the part to go back to.
Moving on marks this lesson complete. Finish the build first, it is the part that counts.